Anthony Piotrowski

– Shipped an analyst-facing retrieval-augmented generation (RAG) agent used daily by 100+ equity researchers, cutting coverage ramp-up from weeks to days; owned the full architecture: SEC filing parsing (10-K/10-Q/8-K), semantic chunking, metadata strategy, vector retrieval tuning, and prompt engineering for high-precision, citation-backed financial answers across hundreds of thousands of documents. – Built an agentic orchestration layer using LangGraph for multi-step financial research workflows — tool-use across retrieval, calculation, and summarization steps — with output guardrails to enforce structure and reduce hallucination risk in production. – Developed an in-house LLM evaluation framework (LangSmith) establishing benchmarks for accuracy, latency, and cost across semi-structured and unstructured financial data tasks; framework gates all LLM feature releases to production. – Extended LLM evaluation work to include fine-tuning experiments — adapted Mistral via LoRA/QLoRA for style and tone transfer, running full pipeline from data curation through eval and deployment to benchmark fine-tuned vs. prompted baselines. – Engineered a systematic DCF valuation engine covering 40,000+ global stocks using regression-based parameter estimation; validated outputs against analyst consensus to scale automated coverage with minimal analyst intervention. – Built product-facing analytics dashboards (Streamlit, Plotly Dash) surfacing portfolio metrics, valuation comparables, and earnings trends used in client-facing investment products.